NVIDIA Online GTC 2020 Kicks Off Today But No Open-Source Linux Announcement Expected

[CommunityMember]

Yeah, maybe you're right. They created some RISC-V based controller called Falcon that they use on their graphics cards and they joined the RISC-V Consortium. I think they might have paid money to join too.

Falcon was originally a custom proprietary design "FAst Logic CONtroller" doing some mundane tasks which used a proprietary toolchain targeting their Falcon design. Around four years ago nvidia announced their next generation would be based on RISC-V, which would give them a bit more performance and capabilities while being reasonably cheap and extensible (no ISA/architecture licensing required) and also allowed them to use a more normal toolchain.

Being part of the RISC-V consortium (and the cost is pocket change to a company like nvidia) gives them a seat at the table to ensure that future directions and changes are not a negative to nvidias interests. Lots of companies like to keep their options open regarding future opportunities, and nvidia certainly wanted to be able to sell GPUs into a RISC-V solution (nvidia was taking the long view).

[MadeUpName]

Once upon a time there was a stream a frog and a scorpion ...

[JPFSanders]

It would be disappointing if they burried such plans if they had existed at some point in the past. But on the other hand it seems to be ineviteable after all that Nvidia one day also open sources major parts of their stack or both AMD and Intel could play that card to their advantage in the future.

Maybe they haven't something to show right now, or it is not the right time to even talk about it - but with all the investments of Red Hat, the recent ARM deal, I guess they are working behind the scenes on such a project.

At this point the NVidia situation feels like deliberated sabotage more than anything else.

We do not need NVidia to open source anything, just allow re-distribution of the signed firmware and some specs.

But it seems they want to avoid MESA to run on their cards at all costs.

This situation with NVidia doesn't make any sense at all.

[DanL]

I have been waiting for nVidia to release the f**king firmware for too long already. Recently I couldn't wait anymore.

Well, the oldest cards to not have firmware (Maxwell) are still included in Nvidia's current driver. Even after they move to legacy branch, they'll still be supported for many years. Not sure why you "couldn't wait anymore".

[ms178]

Maybe, but life taught me time is important. I have been waiting for nVidia to release the f**king firmware for too long already. Recently I couldn't wait anymore, I needed new hardware and I casted my vote with my wallet. My wallet isn't going to fill up again for a long time, so nVidia has lost me for who knows how long. Maybe forever. Maybe I'll be just fine with AMD and I won't feel the desire to buy nVidia ever again. Maybe when I'll need new hardware again, nVidia won't even be a thing anymore. Let's hope it dies in fire.

You couldn't wait until the new generation of both companies is on the market? But I know that feeling, I was desperate enough to build two weeks before Zen 2 launched last year myself.

[lucrus]

Well, the oldest cards to not have firmware (Maxwell) are still included in Nvidia's current driver. Even after they move to legacy branch, they'll still be supported for many years. Not sure why you "couldn't wait anymore".

I couldn't wait anymore to be sure that, if my notebook hangs hard under GPU load (which was happening quite frequently), it's not because of some proprietary nvidia bits that taint the kernel and that no one is willing to debug (rightfully).

I couldn't wait anymore to test newer kernels when I want to (I usually test kernels starting from -rc1). I'm not a kernel developer, but I like testing newer kernels, I like trying new kernel features and I like playing games too. I'm not a hardcore gamer, so I don't need to squeeze every possible FPS from my video card, but nouveau falls short even for a casual gamer like me, and nvidia prevented me to test newer kernels.

I couldn't wait anymore to be sure that, when my notebook hangs while resuming from suspend, failing to light up the display, it's not because of the same nvidia bits above.

In the end, I couldn't wait anymore to own the hardware I paid for.

[bug77]

At this point the NVidia situation feels like deliberated sabotage more than anything else.

We do not need NVidia to open source anything, just allow re-distribution of the signed firmware and some specs.

But it seems they want to avoid MESA to run on their cards at all costs.

This situation with NVidia doesn't make any sense at all.

While open source is obviously preferred, the situation makes perfect sense if you remove your head from the sand, take a step back and look at the bigger picture.

[Paradigm Shifter]

I hope this means (renewed?) support for the (older) Jetson Nano. I've been testing an unofficial Ubuntu 20.04 release and it's generally been smooth sailing... but it would be nice to see some official support beyond the 18.04-based CUDA-10-based L4T - it would also be nice to see a newer kernel than 4.9...

It was amusing to compile the whole Cinnamon 4.6.x DE and have it running at 4K, 60Hz on ARM64 and most importantly... smoothly! Sure, it has the occasional stutter if I ask it to do too much at once (a microSD card issue, I think) but its still very usable. I intend to run through the whole thing again to finish up documenting it, then I'm tempted to set up a PPA as there are no ARM64 builds of Cinnamon currently. That said, Cinnamon is unusable on any other ARM SBC, so... YMMV.

That said, I do question how much point there is to a 2GB Jetson Nano for "AI"... since the CUDA apps I've used and/or coded struggle enough with 4GB...

[JPFSanders]

While open source is obviously preferred, the situation makes perfect sense if you remove your head from the sand, take a step back and look at the bigger picture.

Let's imagine for a second that I'm very stupid and ignorant, can you explain to a stupid ignorant (me) why NVidia behaves with their Linux drivers in a way and with Tegra in another way?

I'm very lost, help me please.

[bug77]

Let's imagine for a second that I'm very stupid and ignorant, can you explain to a stupid ignorant (me) why NVidia behaves with their Linux drivers in a way and with Tegra in another way?

I'm very lost, help me please.

Different products, different strategies.

Nvidia makes a lot of money having a unified driver across all platforms they support. There's little incentive for them to do away with that to target a few percent of their customers. It's a simple business decision, nothing else.